S******y
发帖数: 1123 | 1 I am using ROCR package in R for binary classification.
I am now using default to get ROC curves (true positive vs. false positive).
How can I set my own cutting points (i.e. if prediction prob > 0.65 then
classify as 1; else 0).
How does ROCR set its default cutting points?
Thanks! |
D******n
发帖数: 2836 | 2 ROC was exactly invented to avoid setting up a cutoff.
Each point of the curve corresponds to one cutoff.
).
【在 S******y 的大作中提到】
: I am using ROCR package in R for binary classification.
: I am now using default to get ROC curves (true positive vs. false positive).
: How can I set my own cutting points (i.e. if prediction prob > 0.65 then
: classify as 1; else 0).
: How does ROCR set its default cutting points?
: Thanks!
|
S******y
发帖数: 1123 | 3 谢谢回复。。。
继续钻研中。。。
觉得可能要读ROCR的source code 才能完全明白。。。
我的问题原于这个example-
http://heuristically.wordpress.com/2009/12/23/compare-performance-machine-learning-classifiers-r/
记得好像自己以前用过SVM,KNN得到的是predicted probabilities. 需要自己决定cut
off points... |
